Almost Everywhere

In measure theory, almost everywhere describes a property that holds at every point except possibly within a null set, a set of measure zero. Usually abbreviated a.e., it provides a precise way to disregard exceptional points without claiming that no exceptions exist. The meaning depends on the measure under consideration: a set negligible for one measure may not be negligible for another. Almost-everywhere statements are fundamental to integration, convergence, and the study of spaces of functions. (kmperfekt.com)

Definition and exceptional sets

Let $(X,\Sigma,\mu)$ be a measure space, where $\Sigma$ is a sigma-algebra of measurable subsets of $X$. A property $P(x)$ holds $\mu$-almost everywhere if there exists $N\in\Sigma$ with $\mu(N)=0$ such that $P(x)$ holds for every $x\in X\setminus N$. If the exceptional set

is measurable, this is equivalent to $\mu(E)=0$. The containing-null-set formulation also accommodates exceptional sets that are not themselves measurable. (kmperfekt.com)

A countable union of null sets remains null. Consequently, countably many properties, each holding almost everywhere, hold simultaneously outside one null set. This does not extend to arbitrary uncountable families. For example, for each $t\in[0,1]$, the property $x\ne t$ holds almost everywhere for Lebesgue measure; nevertheless, no $x\in[0,1]$ satisfies it for every $t$. The distinction follows from the fact that the union of all these exceptional singletons is the entire interval. (kmperfekt.com)

Examples and dependence on the measure

For Lebesgue measure on the real line, every singleton has measure zero. Thus every countable set, including the rational numbers, is null. It follows that almost every real number is irrational, even though the rationals form a dense set: every nonempty open interval contains rational numbers. Measure-theoretic negligibility therefore does not mean that exceptions are isolated or absent from small neighborhoods. (math.ucdavis.edu)

Consider the function $f(x)=1$ for rational $x$ and $f(x)=0$ otherwise. It equals zero almost everywhere, although it is discontinuous at every point. This illustrates that almost-everywhere equality need not preserve continuity. (math.ucdavis.edu)

By contrast, under counting measure, every nonempty measurable set has positive measure, so “almost everywhere” means “everywhere.” Under a point mass concentrated at $a$, a property holds almost everywhere precisely when it holds at $a$, regardless of its behavior elsewhere. These conclusions follow directly from the respective measures’ definitions. (kmperfekt.com)

Equality and integration

Two measurable functions are equal almost everywhere when

This defines an equivalence relation, rather than ordinary pointwise equality. For integrable functions, changing values on a null set does not change the Lebesgue integral. Thus if $f=g$ almost everywhere and $g$ is integrable, then $f$ is integrable and their integrals agree. (kmperfekt.com)

The $L^p$ spaces formalize this identification: their elements are equivalence classes of functions equal almost everywhere. For $1\le p<\infty$,

This expression vanishes whenever $f=0$ almost everywhere, even if some individual values are nonzero. Identifying such functions makes it a genuine norm. With this convention, $L^p$ is a Banach space, and $L^2$ is a Hilbert space. (math.ucdavis.edu)

Measurability requires care: arbitrary changes on a null set preserve measurability on a complete measure space, where every subset of a null set is measurable. They need not do so on an incomplete space. (math.ucdavis.edu)

Almost-everywhere convergence

A sequence $f_n$ converges to $f$ almost everywhere when

for all $x$ outside one null set. It is therefore pointwise convergence with a negligible exceptional set, not a statement about convergence at every point or about uniform rates. (kmperfekt.com)

On a finite measure space, almost-everywhere convergence implies convergence in measure, meaning that for every $\varepsilon>0$,

The converse generally fails for the full sequence, but convergence in measure guarantees a subsequence converging almost everywhere. Egorov’s theorem gives another connection: for finite measure and real-valued measurable functions, almost-everywhere convergence becomes uniform convergence after removing a measurable set of arbitrarily small measure. This removed set need not have measure zero. (heil.math.gatech.edu)

Almost-everywhere convergence alone does not justify exchanging a limit and an integral. For example, on $(0,1)$, let $f_n=n\mathbf1_{(0,1/n)}$, where $\mathbf1$ denotes the indicator function. Then $f_n\to0$ at every point, but $\int_0^1 f_n\,dx=1$. The dominated convergence theorem supplies sufficient additional hypotheses: if $|f_n|\le g$ almost everywhere for an integrable $g$, and $f_n\to f$ almost everywhere, then the integrals converge to the integral of $f$. (math.ucdavis.edu)

Probability terminology

On a probability space, “almost everywhere” with respect to the probability measure is called almost surely. For random variables, almost-sure convergence means

The exceptional outcomes have probability zero; they need not be nonexistent. Likewise, equality almost surely identifies random variables that differ only on such outcomes, the probability-theoretic version of almost-everywhere equality. (math.ucdavis.edu)